Line shopping saves money the same way low spreads save traders. Automating the routine turns “I’ll check later” into repeatable, low-latency decisions. The goal is simple: surface the best available price within your limits, then execute with discipline.
Define goals and constraints
Decide what you’re optimizing: best price at time of bet, closing line value, or a minimum expected edge. Pick one as the north star. Everything else falls in line, including when you skip.
Write down constraints before you script. Which books can you actually bet, what are the stake caps, and how fast can you place? Automation should respect limits so alerts are actionable, not aspirational.
Scope your markets
Start narrow. Focus on sides and totals in leagues you can bet daily, then add props once your pipeline is stable. Breadth without reliability creates noise and missed windows.
Define your time zones and lock times. A routine that misses a 1:00 PM kick because it runs on UTC is not a routine; it’s drift. Normalize all alerts to your local clock and show time-to-lock.
Data and alerts pipeline
Use two feeds: one aggregated odds source for breadth and one or two direct-book APIs or parsers you actually bet. Aggregate for discovery; verify on the book you’ll click. This avoids chasing stale numbers.
Convert all odds to implied probability, remove vig, and compute edge vs your fair line or a consensus anchor. Alert only when edge exceeds a threshold and stake is available, or you’ll learn to ignore pings.
Alert hygiene
Keep alerts atomic: one market, one side, one stake suggestion. If you bundle, you delay decisions hunting for details. Latency kills value more often than a slightly imperfect model.
Log why the alert fired. Store snapshot prices across books, your fair, edge after vig, and cap logic. When you review, you’ll fix rules instead of guessing.
Daily workflow: morning, pre-game, live
Start with a morning sweep. Sync limits, verify login state, and refresh your exclusion list for books that throttled you. Clear stale markets and archive yesterday’s logs for review.
Run tighter scans as lock approaches. Increase frequency 60 minutes out, then again 15 minutes out. Decide stake tiers beforehand so you click, not think.
| Window | Cadence | Actions | Notes |
|---|---|---|---|
| Morning | Every 30–60 min | Sync limits, sanity-check model | Disable broken feeds |
| Midday | Every 10–15 min | Hunt soft opens, set feelers | Small stakes only |
| 60–30 min | Every 3–5 min | Main entries per edge rules | Verify stake available |
| 15–0 min | Every 30–60 sec | Top-ups, last looks | Don’t move lines for pennies |
| In-play breaks | Immediate | Halftime/timeout scans | Prebuilt queries only |
Execution rules and guardrails
Automate stake calculation with tiers. For example: 0.5 units at 1.0–1.4% edge, 0.75 units at 1.5–1.9%, 1.0 unit at ≥2.0%, capped by book limit. Fixed tiers prevent overfitting edge decimals.
Respect a minimum actionable price improvement. If switching books saves less than your friction cost (time, risk of limit, withdrawal fees), skip. Don’t pay a dollar to save a dime.
Stake sizing tiers
Tie unit size to total daily risk, not just per-bet edge. A strong slate can still overexpose you if alerts cluster near lock. Cap total exposure per league and per hour block.
If a book slashes limits after a hit, drop it into a cool-down list automatically. Your robot shouldn’t nag you to chase crumbs that trigger more throttling.
Post-mortem and tuning
Judge decisions by the price you captured, not just outcomes. Track closing line value against your entry and against the best available in your feed at the same second. This tells you if the routine is fast enough.
Prune rules monthly. Kill low-yield markets, raise edge thresholds where fills are scarce, and retire books that underdeliver after friction. Simplicity survives; complexity drifts.
Quick checklist (run daily):
- Refresh limits, sessions, and time sync.
- Verify feeds; alert only on edge after vig and stake ≥ minimum.
- Apply stake tiers and exposure caps automatically.
- Log captured price, timestamp, and book; compare to close.
- Review misses and adjust cadence or thresholds, not vibes.